-Author name in bold denotes the presenting author
-Asterisk * with author name denotes a Non-ASH member
Clinically Relevant Abstract denotes an abstract that is clinically relevant.

PhD Trainee denotes that this is a recommended PHD Trainee Session.

Ticketed Session denotes that this is a ticketed session.

4983 Application of a Novel Artificial Intelligence Algorithm to Understand Spatial Immune Cell Relationships in Newly Diagnosed Multiple Myeloma

Program: Oral and Poster Abstracts
Session: 803. Emerging Tools, Techniques, and Artificial Intelligence in Hematology: Poster III
Hematology Disease Topics & Pathways:
Research, Artificial intelligence (AI), Adult, Clinical Practice (Health Services and Quality), Clinical Research, Plasma Cell Disorders, Diseases, Immune mechanism, Lymphoid Malignancies, Biological Processes, Emerging technologies, Technology and Procedures, Study Population, Human, Pathogenesis, Machine learning, Pathology
Monday, December 9, 2024, 6:00 PM-8:00 PM

Paraic Behan1*, Sahin Sarihan, BSc1*, Megan Chiasson1*, Pir Zainulabdeen Jan Sarhandi1*, Joanna Fay2*, John P Quinn3*, Philip T Murphy3*, Katherine M Sheehan2* and Siobhan V Glavey1

1Departments of Haematology and Pathology, Beaumont RCSI Cancer Centre, Dublin, Ireland
2Department of Pathology, Beaumont RCSI Cancer Centre, Dublin, Ireland
3Department of Haematology, Beaumont RCSI Cancer Centre, Dublin, Ireland

Background:

The unique features of the implicated plasma cell clone and its interaction with the tumour microenvironment are integral to the understanding of the pathophysiology of multiple myeloma (MM). It has been postulated that the colocalization of bone marrow plasma cells (BMPCs) and other immune cells, such as eosinophils, in the bone marrow may provide a permissive niche for BMPC survival via APRIL and IL-6 production1. To understand if spatial relationships of cellular components within the bone marrow microenvironment in newly diagnosed multiple myeloma (ND-MM) are functionally relevant in patients, we combined whole slide imaging (WSI) with a novel artificial intelligence (AI) and machine learning algorithm to assess cellular proximity, spatial relationships and plasma cell density.

Methods:

Bone marrow (BM) trephine WSIs were retrospectively analysed from 29 treatment naïve, ND-MM patients. WSIs were created by scanning the H&E trephine slides at X40 magnification using a high-power digital scanner (Objective Imaging). The HALO AI (Indica Labs, NM, USA) DenseNet v2 tissue classifier was trained to segregate bone, cellular and adipose tissue compartments, followed by quantification of BMPCs and eosinophils (nuclear phenotyping). The HALO Spatial analysis module was then used to assess the distribution of BMPCs within the marrow cellular compartment (spatial analysis) and the proximity of eosinophils to BMPCs (proximity analysis). Clinical biomarkers (e.g. Beta-2-microglobulin [B2M], LDH, peripheral white cell counts) were compared with AI generated data examining the spatial relationships between BMPCs and key immunological elements of the bone marrow microenvironment.

Results:

The median age of the 29 ND-MM patients was 66 years with a balanced male to female ratio (52% and 48% respectively). Based on cytogenetic and clinical data, 90% of patients were stratified as having IMWG R-ISS standard-risk disease and the remaining 10% were stratified as having high-risk MM. The tissue classification analysis by AI correctly classified ND-MM in all 29 cases by establishing mean BMPC number, density BMPCs in mm2 and the total BMPC percentage. This was possible on H&E stained slides and did not require immunohistochemistry for CD138. There was positive diagnostic concordance (R = 0.56, p<0.01) between nuclear phenotyping by AI on H&E slides and the BMPC percentage on CD138 stained slides as estimated by an independent histopathologist. Spatial analysis demonstrated that increasing BMPC density at incremental distances from the bony trabecula positively correlated with B2M. Incremental B2M was positively correlated with both BMPC density in mm2 (R = 0.46, p<0.05) and BMPC distance from the bony interface to the deeper bone marrow compartment of 120 to 300μm (R = 0.5-0.61, p<0.05) indicating that B2M may predict invasive behaviour and homotypic interactions of BMPCs. Proximity analysis demonstrated an increased intercellular distance between eosinophils and BMPCs in kappa restricted MM (R = 0.74-0.8, p<0.001) which may indicate altered immune cell cross-talk in this subtype of MM. The absolute neutrophil count in the peripheral blood was found to be correlated with an intermediate distance between eosinophils and plasma cells of 100-140μm (R = 0.61-0.65 p<0.001) indicating that PC trafficking may be altered in patients with neutrophilia at diagnosis.

Conclusion:

Our AI algorithm can robustly and correctly identify MM from BM trephine H&E WSI, negating the need for CD138 staining, which may facilitate earlier diagnostic certainty. Additionally, our AI analysis indicates that immune cell-PC spatial relationships within the bone marrow microenvironment are altered in high disease burden states (higher B2M) and are impacted by the clonal subtype of MM and neutrophil-eosinophil localisation, perhaps indicating a pro-inflammatory milieu in the bone marrow. More work is needed to assess if AI imaging at diagnosis may in the future aid risk stratification and therapy selection for ND-MM patients.

References:

1. Chu VT, Beller A, Rausch S, Strandmark J, Zanker M, Arbach O, et al. Eosinophils promote generation and maintenance of immunoglobulin-A-expressing plasma cells and contribute to gut immune homeostasis. (2014) 40:582–93. 10.1016/j.immuni.2014.02.014

Disclosures: Glavey: Skyline Dx: Research Funding; Pfizer: Honoraria, Research Funding; ReNAgade: Consultancy; Amgen: Honoraria, Research Funding, Speakers Bureau; Janssen: Honoraria, Research Funding, Speakers Bureau.

*signifies non-member of ASH